ASD ISM 2093 Role-based access controls in AI applications
What it says
Role-based access controls are implemented for AI applications to restrict access to sensitive data.
Quoted as published. Everything else on this page is written by Control Stack.
What it means
In plain EnglishAccess to sensitive data in AI apps is restricted based on user roles.
This control ensures that people can only access sensitive information in AI applications if they have a role that needs it. Without these restrictions, sensitive data could fall into the wrong hands, leading to privacy breaches and potential business losses.
Free. Save your status, tasks and notes.
Why it matters
If role-based access controls aren't used, sensitive data could be accessed by unauthorised users, potentially leading to data breaches and financial loss.
How to implement
Managers should assign roles carefully: Identify the employees who need access to sensitive data within AI applications based on their job responsibilities.
Document their roles and ensure they are updated as responsibilities change.
IT teams should set up user accounts: Configure the software to link each user account to their assigned role, which dictates the level of access to data they have.
Use software tools that allow for easy role-based access settings.
HR should train employees on access policies: Provide clear guidelines to staff about data access and privacy rules relevant to their roles.
Conduct regular training to reinforce these policies and ensure understanding.
System owners should review access permissions periodically: Schedule regular checks (at least quarterly) to ensure that who has access to sensitive data aligns with current job roles and tasks.
Adjust roles as job functions evolve.
Security teams should audit access logs: Use software to track and monitor who accesses what information and when, allowing for quick detection of any unauthorised access attempts.
Keep it running
Regularly review and update user roles to accommodate job changes or personnel shifts, ensuring access to sensitive data remains appropriate.
Evidence and audit
- Check 1 of 5What an auditor asks for
Ask for
Ask for the role-based access control policy document covering the organisation's AI applications.
Look at
Check whether the document outlines how roles are assigned and how access to sensitive data is managed.
Good evidence
The policy details roles, responsibilities and the process for changing access.
- Check 2 of 5What an auditor asks for
Ask for
Request a list of current user access roles for the AI applications.
Look at
Verify that each user matches the roles outlined in the policy and that outdated roles are not active.
Good evidence
User roles align correctly with the policy, with minimal discrepancies.
- Check 3 of 5What an auditor asks for
Ask for
Ask to see recent access log reports for the AI applications.
Look at
Review the logs for entries showing anomalous access attempts or misuse of access rights.
Good evidence
Logs show consistent, authorised access patterns with no irregular activity.
- Check 4 of 5What an auditor asks for
Ask for
Request documentation of recent role reviews.
Look at
Check for records of when access roles were last reviewed and updated.
Good evidence
There is documented evidence of regular reviews with clear notes on the changes made.
- Check 5 of 5What an auditor asks for
Ask for
Ask for training records on the access policy.
Look at
Review employee training completion records to confirm staff received instruction on the access policy.
Good evidence
All relevant staff have completed up-to-date training on the access policy.
Mappings
Closest match in ISO/IEC 27001: A.5.15, Access control policies and procedures, which this control partially meets.
ISO/IEC 27001: 3 mappings
This control partially meets A.5.15.
ISM-2093 requires organisations to implement RBAC in AI applications to prevent unauthorised access to sensitive AI data. Annex A 5.15 requires establishing and implementing rules and procedures to control logical (and physical) access to information assets based on business and security requirements. ISM-2093 fits within these broader access control rules as a specific procedural/technical requirement for AI systems.
This control partially meets A.8.3.
ISM-2093 requires RBAC for AI applications to restrict access to sensitive AI data. Annex A 8.3 requires restricting access to information and associated assets according to an access control policy, which includes AI data as a subset. ISM-2093 is a specific, AI-focused instantiation of the broader ISO requirement to restrict access based on policy.
This control partially overlaps A.5.18.
ISM-2093 requires RBAC enforcement in AI applications so only authorised roles can access sensitive AI data. Annex A 5.18 addresses the access rights lifecycle (provisioning, review, modification and removal) for information assets, which is how RBAC entitlements for AI data are administered. They overlap on controlling who has access, but Annex A 5.18 emphasises governance and ongoing reviews whereas ISM-2093 emphasises implementing RBAC restrictions in the AI application itself.
ASD Essential Eight: 1 mapping
This control partially overlaps E8-RA-ML3.1.
ISM-2093 requires role-based access controls (RBAC) for AI applications to restrict access to sensitive AI data to authorised personnel. E8-RA-ML3.1 aligns by requiring privileged access to systems, applications and data repositories to be limited to what is necessary for duties, which commonly uses role-based entitlements. However, E8-RA-ML3.1 is focused specifically on privileged access across the environment, while ISM-2093 is focused on AI application data access (privileged and non-privileged) using RBAC.
Mappings show how controls relate across frameworks. They do not imply full equivalence or certification.